Multiresolution image alignment based on discrete wavelet transform

We introduce a multi-resolution image registration based on using discrete wavelet transform. We first extract contour from both images that we want to align. The extracted contours are then fitted with B-spline curve representation to synthesize the new contours with equal number of point. The area parameter is used in the B-spline fitting to make the new generated curve immune to affine transformation. Before representing the B-spline contour with discrete wavelet transform, the problem of starting point of the contour needs to be handle. This can be done by computing the maximum curvature. The maximum curvature is selected as the starting point. Once the starting points on the contour have been established, the discrete wavelet transform is then recursively represented the contours until only a few points are remained. Due to the affine-invariant properties of discrete wavelet transform, these points can be used as landmark points for registering the transformed contour with the original contour. The experiments have shown that the purposed methods are robust and promising even in the presence of noise.